Prominence calculator from findpeaks seems off?

Hi all,

I'm trying out the findpeaks function to find the most prominent peak. The function documentation (see here) gives a pretty clear explanation of how the prominence is calculator. In short, from the central peak, find the lowest point to the left and right of said peak, and take the distance from the highest of these two untill the peak (someone correct me if I'm wrong here). However, I'm not sure if I'm missing a point or there is something wrong here.

First figure

For the first signal (first figure below), this seems to be in the correct ballpark. Very roughly, I would say this is the distance between the low point (reference point) around -9 (just over x=5000), and the peak (let's say 12), which would give a prominence of about 21. The reported prominence by the function is 27.94. First of all, this doesn't seem to fit entirely with the ballpark estimation, but this was very rough.

Second figure

Now looking at the second figure, the positive peak here is far less prominent. The low, reference point in this case would be about -2.2 or so (almost touching x=0), and the peak seems to be around 3.2 or so. One would guestimate this prominance at about 5.4 or so. However, the findpeaks function gives a prominence of 65.57.

First of all, this doesn't check out with the reference-to-peak rationale I was following, and second, this prominence value is way higher than the value of the first signal, which does not seems to make sense to me.

I'm using the function as follows, where data is always a 1x10000 vector of singles

findpeaks(data,'NPeaks',1,'SortStr','descend');

I think I understand the rationale of the prominence calculation from the function documentation, but I could be wrong. Otherwise I don't really know what's going on here. If anyone can point me in the right direction, that would be great (Or if anyone wants to verify that this function is behaving weird, that's okay too).

(NB. There are many more other peaks that also don't seem to check out in terms of prominence calculators)

[dpb]

Would have to have the .mat files of the data to have any chance at all of determining what is going on.

[Reinder Dorman]

Added the .mat files for the two examples.

However, I notice the same thing when just using something from rand(). This giving random values between 1 and 0, but on a number of test, the prominence is always above 1 (most values between 1 and 6). This doesn't seem to correspond with (what I think is) the explanation of the prominence value.

[David Goodmanson]

Hello Reinder,

how, exactly, are you using the findpeaks command in order to produce the prominences?

[Reinder Dorman]

Hi David,

I produce the figures as mentioned in the little code snippet above, when catching the prominences (and the other values I basically add the left side, as such:

[peak,offset,prominence] = findpeaks(data,'NPeaks',1,'SortStr','descend');

[Just Manuel]

well, this looked like some usage issue to me :)

If you look at the docs, there are these 3 signatures for output arguments:

pks = findpeaks(data)
[pks,locs] = findpeaks(data)
[pks,locs,w,p] = findpeaks(data)

so, in order to get the prominence, you need to use the third one like this:

[peak, offset, ~, prominence] = findpeaks(...)

cheers

Manuel

[Reinder Dorman]

I knew it. I frikkin knew it. I can hardly express myself now without turning to faul language, so I'll keep my emotion away from this board.

Thanks for this. And it again confirms the old credo "the computer is never wrong, the programmer is".